Application of machine/deep-learning to the systems biology of glycosylation

NHLBI资助的“糖基化的系统生物学”旨在将生物分子工程方法应用于 从基础科学和转化的角度研究血细胞糖基化。目标是开发一个 细胞转录组和表观遗传状态之间的定量联系，以及由此产生的糖基化谱。 该基金的一部分重点是发现血细胞中的细胞调节途径， 这些细胞上糖基化的模式，并评估这些原则的可推广程度。在 第二个方面，利用这些新知识，我们确定测量选定的遗传特征是否可以报告 对细胞糖基化状态的影响。这些关键标记/检查点的识别具有翻译性 重要性，因为它可以在临床试验和精准医学的背景下为患者分层提供信息 应用.为了实现这些目标，两种类型的扰动实验进行使用 不同的血细胞首先，CRISPR-Cas9/gRNA用于实现定义的系统扰动， 测量细胞糖组中产生的变化。这表示来自机器的“标记数据集” 学习/深度学习（ML/DL）视角。在第二种方法中，施加生化刺激以扰乱细胞状态， 并再次进行细胞糖基化状态测量。这是作为扰动的“未标记数据集”， 是不精确的。在每种情况下，测量几个实验输出或“特征”，包括： 代测序（NGS），用于同时定量基础转录组、gRNA的性质 （向导-RNA）扰动和糖基化状态（使用凝集素结合）。2)谱流 流式细胞术，以测量荧光凝集素结合在更大的规模，与选择的稀有细胞类型， 进行更深入的分析3)质谱法以获得详细的聚糖结构数据。数学 开发方法来融合来自这些不同的组学方法的结果并开发输入-输出响应。 目前，这样的建模依赖于先前的生化知识，这些知识在途径图中进行管理，线性混合 模型和显式编程。作为这种传统方法的替代方法，本补编将编写 ML/DL建模和相关学习的数据。为了实现这一点，我们增加了两个新的专家调查员， Gunawan（系统生物学，单细胞分析）和Chen（机器/深度学习）。具体目标 将：1）为ML/DL收集足够的数据; 2）规范和标准化这些数据，以便ML/DL准备就绪;以及3） 在试点ML/DL测试中使用转换后的数据。项目的成功完成将确认ML/DL的价值， 研究血细胞和糖科学应用。据我们所知，这将是第一个 ML/DL在多组学糖科学中的应用与传统建模方法相比， 已经在资助的应用程序中得到支持，将告诉我们ML/DL的优点和局限性。最后 一个通用的ML/DL数据处理框架将出现，可以应用于本项目的其他方面， 以及其他相关的生物医学问题。

The NHLBI grant “Systems Biology of Glycosylation” aims to apply biomolecular engineering approaches to study blood cell glycosylation from both a basic science and translational perspective. The goal is to develop a quantitative link between the cellular transcriptome and epigenetic status, with the resulting glycosylation profile. A portion of the grant is focused on discovering the cellular regulatory pathways in blood cells that control the pattern of glycosylation on these cells, and assessing the extent to which these principles are generalizable. In a second aspect, using this new knowledge, we determine if measuring selected genetic signatures can report on the glycosylation status of cells. The identification of these key makers/checkpoints has translational significance as it can inform both patient stratification in the context of clinical trials and precision medicine applications. In order to achieve these objectives, two types of perturbation experiments are performed using different blood cells. In the first, CRISPR-Cas9/gRNA is used to implement defined system perturbations and resulting changes in the cellular glycome are measured. This represents the ‘labeled dataset’ from the Machine Learning/Deep Learning (ML/DL) perspective. In the second, biochemical stimuli are applied to perturb cell state, and again cell glycosylation status measurements are made. This is the ’unlabeled dataset’ as the perturbation is imprecise. In each case several experimental outputs or ‘features’ are measured including: 1) Single-cell next- generation sequencing (NGS) for the simultaneous quantitation of the underlying transcriptome, nature of gRNA (guide-RNA) perturbation and glycosylation status (using lectin binding), on individual cells. 2) Spectral flow cytometry to measure fluorescent lectin binding in larger scale, with the option that selected rare cell types could be sorted for more in-depth profiling. 3) Mass spectrometry to obtain detailed glycan structure data. Mathematical methods are developed to fuse results from these different omics-methods and develop input-output responses. Currently, such modeling relies on prior biochemical knowledge that is curated in pathway maps, linear-mixed models and explicit programming. As an alternative to this traditional approach, this supplement will prepare the data for ML/DL modeling and related learning. To achieve this, we add two new expert investigators to this project: Gunawan (systems biology, single-cell analysis) and Chen (machine/deep learning). The specific aims will: 1) Collect sufficient data for ML/DL; 2) Normalize and standardize these data for ML/DL readiness; and 3) Use the transformed data in pilot ML/DL tests. Successful project completion will confirm the value of ML/DL in the study of blood cell and Glycoscience applications. To our best knowledge, this would represent the first application of ML/DL to multi-omics Glycosciences. A comparison with traditional modeling methods that are already supported in the funded application, will tell us about the merits and limitations of ML/DL. Finally, a general ML/DL data processing framework will emerge that can be applied to other aspects of this project and also other related biomedical problems.